diff options
| author | Alexander Potapenko <glider@google.com> | 2026-01-30 11:16:57 +0100 |
|---|---|---|
| committer | Alexander Potapenko <glider@google.com> | 2026-02-09 09:43:36 +0000 |
| commit | eababac62cc8b2dbcdadba991eecaa4fd521c42d (patch) | |
| tree | a097456de05dd7dc770dd539292cb527ba3fc43b /sys | |
| parent | d733ce389c3287808dd15952c6aa481021110196 (diff) | |
sys/linux/test: add amd64-syz_kvm_nested_load_syzos
The new test checks the following scenario:
- L0 VM launches L1 VM (SYZOS)
- L1 VM launches L2 VM (nested SYZOS)
- L2 VM issues a UEXIT
- L1 VM catches the UEXIT and forwards it to L0
- L0 VM checks the UEXIT code and resumes L1 VM
- L1 VM finishes its work and issues another UEXIT
- L0 VM checks the UEXIT code
Diffstat (limited to 'sys')
| -rw-r--r-- | sys/linux/test/amd64-syz_kvm_nested_load_syzos | 25 |
1 files changed, 25 insertions, 0 deletions
diff --git a/sys/linux/test/amd64-syz_kvm_nested_load_syzos b/sys/linux/test/amd64-syz_kvm_nested_load_syzos new file mode 100644 index 000000000..564680952 --- /dev/null +++ b/sys/linux/test/amd64-syz_kvm_nested_load_syzos @@ -0,0 +1,25 @@ +# requires: arch=amd64 -threaded +# +r0 = openat$kvm(0, &AUTO='/dev/kvm\x00', 0x0, 0x0) +r1 = ioctl$KVM_CREATE_VM(r0, AUTO, 0x0) +r2 = syz_kvm_setup_syzos_vm$x86(r1, &(0x7f0000c00000/0x400000)=nil) + +# Create a nested VM that executes a SYZOS uexit payload. +# +r3 = syz_kvm_add_vcpu$x86(r2, &AUTO={0x0, &AUTO=[@enable_nested={AUTO, AUTO, 0x0}, @nested_create_vm={AUTO, AUTO, 0x0}, @nested_load_syzos={AUTO, AUTO, {0x0, 0x10, [@uexit={AUTO, AUTO, 0xFACEFEED}]}}, @nested_vmlaunch={AUTO, AUTO, 0x0}], AUTO}) +r4 = ioctl$KVM_GET_VCPU_MMAP_SIZE(r0, AUTO) +r5 = mmap$KVM_VCPU(&(0x7f0000009000/0x1000)=nil, r4, 0x3, 0x1, r3, 0x0) + +# Run VCPU. L2 should start, execute uexit(0xFACEFEED). +# L1 intercepts this, adds nesting level 1 to the top byte, and re-throws to L0. +# Expected value: 0x01 (Level) << 56 | 0xFACEFEED = 0x01000000FACEFEED +# +ioctl$KVM_RUN(r3, AUTO, 0x0) +syz_kvm_assert_syzos_uexit$x86(r3, r5, 0x01000000FACEFEED) + +# guest_main should finish with guest_uexit(-1). +# L1 executes this directly, so nesting level is 0. +# Expected value: 0x00 << 56 | -1 = 0xFFFFFFFFFFFFFFFF +# +ioctl$KVM_RUN(r3, AUTO, 0x0) +syz_kvm_assert_syzos_uexit$x86(r3, r5, 0xffffffffffffffff) |